behavioral/systems/cognitive alpha

Behavioral/Systems/Cognitive

Alpha/Beta Oscillations Indicate Inhibition of InterferingVisual Memories

Gerd T. Waldhauser,1 Mikael Johansson,1 and Simon Hanslmayr2,3

1Department of Psychology, Lund University, 221 00 Lund, Sweden, and 2Department of Psychology and 3Zukunftskolleg, University of Konstanz, 78457Konstanz, Germany

Selective retrieval of a specific target memory often leads to the forgetting of related but irrelevant memories. Current cognitive theorystates that such retrieval-induced forgetting arises due to inhibition of competing memory traces. To date, however, direct neuralevidence for this claim has not been forthcoming. Studies on selective attention suggest that cortical inhibition is mediated by increasedbrain oscillatory activity in the alpha/beta frequency band. The present study, testing 18 human subjects, investigated whether thesemechanisms can be generalized to selective memory retrieval in which competing memories interfere with the retrieval of a targetmemory. Our experiment was designed so that each cue used to search memory was associated with a target memory and a competitormemory stored in separate brain hemispheres. Retrieval-induced forgetting was observed in a condition in which the competitor memoryinterfered with target retrieval. Increased oscillatory alpha/beta power was observed over the hemisphere housing the sensory represen-tation of the competitor memory trace and predicted the amount of retrieval-induced forgetting in the subsequent memory test. Theseresults provide the first direct evidence for inhibition of competing memories during episodic memory retrieval and suggest thatcompetitive retrieval is governed by inhibitory mechanisms similar to those employed in selective attention.

IntroductionRemembering a previously experienced episode is often challengedby interference from currently irrelevant memory representations.Executive control mechanisms aid selective remembering in suchambiguous retrieval situations by actively inhibiting competingmemories (for review, see Levy and Anderson, 2002). The effect ofinhibition can be observed on a later test, where previously interfer-ing information shows impaired recall, a phenomenon termedretrieval-induced forgetting (RIF; Anderson et al., 1994). Previousneuroimaging studies indicate that prefrontal control mechanismsplay a crucial role in producing RIF (Johansson et al., 2007; Kuhl etal., 2007; Wimber et al., 2009; Hanslmayr et al., 2010; Staudigl et al.,2010). However, no study has yet been able to directly track theinhibition of competing memories due to a spatiotemporal overlapbetween target-related and competitor-related neural activity. Weapplied a visual half-field manipulation in an episodic memory task(Gratton et al., 1997) to disentangle target from competitor-relatedactivity during selective memory retrieval. Our results provide thefirst direct evidence for the inhibition of competing memories.

Participants encoded cue-associate (shape–color) pairs, attend-ing to the cue at central fixation. Cues were presented twice, first

together with an associate in one visual field (e.g., left of fixation) andlater in the sequence of trials with a second associate in the oppositevisual field (e.g., right of fixation; Fig. 1). The two associates eitherhad the same color (non-interference condition) or two differentcolors (interference condition). During a subsequent selective re-trieval phase, some of the cues were presented and participants wereasked to retrieve the target associate studied in one of the visual fields(Fig. 1). The other color associated with the cue, studied in the op-posite visual field, was not addressed and served as the competitorassociate. In a final recall task, memory for all cue-associate pairs wasassessed. RIF was expected for competitors from the interferencecondition encoded with cues that underwent selective retrieval, ascompared to baseline items (cue-associate pairs for which the cuewas not presented during selective retrieval).

EEG analyses contrasted interference and non-interference con-ditions during selective retrieval. Importantly, conditions were iden-tical in terms of sensory stimulation and differed only in whether ornot the competitor associate, encoded contralateral to the target as-sociate, was causing interference (Fig. 2). Studies in the selectivevisual attention domain have shown an increase of alpha/beta power(10–20 Hz) over the brain hemisphere where a competing item isprocessed (Worden et al., 2000; Kelly et al., 2006; Sauseng et al.,2009b; Handel et al., 2011). Additionally, multimodal imaging stud-ies demonstrated that periods of high alpha/beta activity index re-duced excitability of the cortex (Romei et al., 2008; Sauseng et al.,2009a) and decreased metabolic demands (Hanslmayr et al., 2011b;Scheeringa et al., 2011). Therefore, increased alpha/beta power overthe brain hemisphere housing competing memories during selectiveretrieval would signify inhibition (Klimesch et al., 2007; Jensen andMazaheri, 2010; Hanslmayr et al., 2011a). Such a finding would beindicative of a generic, cross-domain mechanism mediating inhibi-tion in both selective visual attention and episodic memory retrieval.

Received Aug. 15, 2011; revised Nov. 9, 2011; accepted Dec. 14, 2011.Author contributions: G.T.W., M.J., and S.H. designed research; G.T.W. performed research; G.T.W. and S.H.

analyzed data; G.T.W., M.J., and S.H. wrote the paper.This work was supported by grants from the Crafoord Foundation (20090712; to G. T. W.), the Swedish Research

Council (VR 60229601; to M. J.), and the German Research Council (DFG HA 5622/1-1; to S.H.). We thank C. Turnstedt,J. Klipvik, and S. Nilsson for assistance with data collection, H. Zimmer for providing the stimulus material, and M.Wimber, B. Spitzer, D. B. Terhune, and T. Staudigl for valuable comments on an earlier version of the article.

Correspondence should be addressed to Gerd Waldhauser, Department of Psychology, Lund University, Box 213,221 00 Lund, Sweden. E-mail: [email protected].

DOI:10.1523/JNEUROSCI.4201-11.2012Copyright © 2012 the authors 0270-6474/12/321953-09$15.00/0

The Journal of Neuroscience, February 8, 2012 • 32(6):1953–1961 • 1953

Materials and MethodsParticipantsEighteen participants (11 female) with a mean age of 25 years (range: 18–38)and normal or corrected-to-normal vision completed the experiment andwere included in the data analysis. All participants were right-handed,reported no history of neurological disease, and gave written informedconsent. The study was approved by the Ethical Review Board at LundUniversity.

Stimulus materialTwo-hundred and sixteen abstract line drawings (Groh-Bordin et al., 2007)were selected and counterbalanced across sessions (108 each), subjects,and conditions. The abstract shapes served as memory cues and werepaired with nine different colors (associates), corresponding to basiccolor categories (blue, green, yellow, red, brown, gray, violet, orange, andpink; Taft and Sivik, 1997). Association between shapes and colors wasrandomized for the first block of the session and subsequently counter-balanced across blocks and shapes so that each color appeared equallyoften in each experimental condition.

Experimental setupParticipants were seated in front of a 17 inch monitor (65 cm dis-tance) with a resolution of 1280 � 1024 pixels. During the experi-ment, the abstract shapes were presented at a central position with asize of 6 � 5 cm (visual angle: 5.3 � 4.4°), together with rectangularboxes, that covered the lower half of the screen (3.1 cm/3° to the leftor right from fixation). The boxes were 9.3 � 13.7 cm in size (vi-sual angle: 8.1 � 11.9°) and were filled with color during the studyphase.

ProcedureParticipants were tested in two sessions on two separate days, 4 –15 dapart (M � 8). Each session contained 18 blocks, with each block includ-ing the whole procedure as described below.

During all phases of the experiment, each trial began with a polychro-matic mask to avoid afterimages from the preceding stimulation (800 msin the study, practice, and final recall phase and 500 ms in the test–feedback and re-study phases). This was followed by a blank screen (300ms), a fixation cross (500 ms), and a stimulation-free baseline interval(1000 –1500 ms in the study, practice, and final recall phase and 200 – 400

Figure 1. Experimental design exemplifying trial procedures for all experimental phases and possible succession of trials from the non-interference (one-color) and interference (two-color) conditions.

Figure 2. Schematic illustration of obtaining the lateralized EEG correlates of inhibition. EEGactivity related to visual stimulation and basic retrieval processing of the target (‘T’) is identicalin the interference and non-interference conditions, whereas inhibition of the competitor (‘C’)in the hemisphere ipsilateral to the retrieval cue is unique to the interference condition (left).

1954 • J. Neurosci., February 8, 2012 • 32(6):1953–1961 Waldhauser et al. • Alpha/Beta Oscillations Indicate Memory Inhibition

ms in the test–feedback and re-study phases). Throughout the experi-ment, participants were instructed to fixate the center of the screen asindicated by the fixation cross or the centrally presented shape and toavoid any eye movements or blinks except during the polychromaticmask. In all phases except for the recall phase, order of presentation wasrandomized, with the constraint that the same shape never occurredtwice in a row.

Study phase. During study, participants were presented with six differ-ent shapes, each combined with instances of all colors. Shapes were pre-sented twice, first together with a color in one visual field and later with acolor presented in the opposite visual field. Half of the shapes were pre-sented with the same color at left and right positions, which comprisedthe non-interference condition. The other half was presented with twodifferent colors, one color at the left position and another color at theright position, which comprised the interference condition. Shapes werepresented for 4000 ms. The color box was flashed for 200 ms at the onsetof the shape to avoid saccades to the color (Gratton et al., 1997). Partic-ipants were instructed to memorize the whole percept including shape,color, and box position in the current trial (left or right). After initialstudy, participants were instructed to count backwards in steps of threefrom a random three digit number for 15 s. Following this distracter task,participants were sequentially presented with all abstract shapes togetherwith a white box to the left or right of fixation and instructed to covertlyrecall the color that was previously presented at that position. After 2000ms of cue presentation, all nine possible colors appeared on the screenand participants were instructed to select the correct color by mouse clickwithin 6000 ms. After a response, participants received feedback on theirperformance on a screen stating “correct” or “wrong” (500 ms). To en-sure high memory performance levels, a re-study phase followed test–feedback recall, which was identical to the initial study block except forthat the shapes were only presented for 1500 ms.

Selective retrieval phase. During the selective retrieval phase, two of theshapes from each interference and non-interference condition were pre-sented. The shapes were displayed for 2000 ms together with one whitebox serving as an unambiguous retrieval cue for the color that was pre-viously presented at that position (target associate). As in previous stud-ies (Hanslmayr et al., 2010; Staudigl et al., 2010), each cue–target pair wasretrieved twice. However, only one position was cued, addressing mem-ory for the target color initially presented in the right visual field (RVF) orin the left visual field (LVF). For each condition, one of the shapes waspresented with a white box in the RVF and the other was presented witha box in the LVF. As in prior studies (Hanslmayr et al., 2010; Staudigl etal., 2010), participants were instructed to covertly retrieve the appropri-ate target color and to try to visualize the initial shape– color combina-tion as it had appeared in the cued visual field during study. Trials wereseparated by 700 ms stimulation-free intertrial-intervals. After selectiveretrieval, a distracter task was carried out (Eriksen flanker task, duration:70 s; Eriksen and Eriksen, 1974; Corballis and Gratton, 2003).

Final recall phase. In the final recall phase, all initially studied color–shape associations were tested. Cue-associate pairs for which the shapewas not presented during the retrieval practice phase served as baselineitems. Associates that were not retrieved, but were associated with thesame shape as the retrieved (target) items, served as competitor items.The test was identical to the recall test during the study phase, with thedifference that the order was constrained such that competitor itemsfrom both interference and non-interference conditions, together with acorresponding set of baseline items, were tested before the target items.This was done to avoid the possibility that output interference in the finalrecall test would confound the effects of memory inhibition (Anderson etal., 1994; Johansson et al., 2007). In addition, presentation of the ninepossible colors was delayed for 500 ms after offset of the retrieval cue.

EEG recording and preprocessingEEG was recorded from 32 Ag/AgCl scalp electrodes with a 500 Hz sam-pling rate and amplified from DC to 200 Hz on a Neuroscan NuAmpssystem, referenced to the left mastoid and re-referenced offline to a com-mon average reference. Data were corrected for eye blinks and verticaleye movements using two additional channels assessing activity in thevertical and horizontal electrooculogram, applying the algorithm imple-

mented in Neuroscan Edit 4.4. Continuous EEG was epoched from �500ms before stimulus onset to 2000 ms poststimulus. All epochs containingremaining artifacts as identified by careful visual inspection were ex-cluded from further analyses. A high-pass filter (0.05 Hz, �3 dB, 12dB/octave roll-off) was applied off-line before further analyses.

Data analysisBehavioral data analyses. To test for RIF, recall performance in thefinal recall test for the competitor items was compared with recallperformance for the baseline items from the interference and the non-interference condition. This was done by means of a 2 � 2 � 2 repeated-measures ANOVA with the factors CONDITION (interference vsnon-interference), ITEM (competitor vs baseline), and HEMISPHERE(left vs right). The last factor was included to test possible hemisphere-specific behavioral effects. The behavioral analysis included only itemsthat were successfully encoded in the study phase as indicated by correctperformance in the recall–feedback test after initial study. RIF effectswere expected to be reflected in a CONDITION � ITEM interaction,characterized by lower recall performance for competitors when com-pared to baseline items in the interference condition. Significant effectswere followed-up by planned comparisons employing uncorrectedpaired-samples Student’s t tests (two-tailed).

EEG analysis methods. Analyses focused on the EEG data recordedduring the selective retrieval phase, where the interference and the non-interference retrieval conditions were contrasted. Note that both condi-tions were identical in terms of sensory stimulation and differed only bywhether or not the competitor item studied contralaterally to the targetitem was interfering (Fig. 2). EEG analysis was carried out using self-written MATLAB (MathWorks) codes applying a Gabor transformationto derive a time–frequency representation from the EEG signal (e.g.,Hanslmayr et al., 2010). To avoid filter artifacts at the edges of the epochs,the data were filtered in a slightly larger time interval, but analyses wererestricted to the 2500 ms epochs. The data were filtered in a frequencyrange of 2–30 Hz with a filter parameter for time–frequency resolution(gamma) set to 1, resulting in a circular spatial aspect ratio of the Gauss-ian function supporting the Gabor transformation. To quantify event-related power changes, poststimulus power change was calculated inrelation to a prestimulus baseline period (�500 to 0 ms; cf. Pfurtschellerand Aranibar, 1977). All statistical analyses of EEG power were con-ducted using nonparametric Wilcoxon signed-rank tests (two-tailed).

Visual field-specific EEG analysis and selection of time windows. Toidentify the time–frequency windows in which hemisphere-specific ef-fects emerged, we subtracted the EEG activity of the non-interferencecondition from the interference condition for trials for which the whiteboxes, functioning as retrieval cues, were presented in the same visualfield during the selective retrieval phase (see Fig. 3). Interference andnon-interference conditions should both entail retrieval processes, butonly the interference condition would contain activity that is related tothe inhibition of competitor items in the hemisphere ipsilateral to theretrieval cue. Thus, the strategy to subtract EEG activity of the non-interference condition from the interference condition should reveal theneural activity that is specific to the effects of retrieval competition andinhibition. The comparison included only trials for which the cue– com-petitor combinations were successfully retrieved during the study phase(Storm et al., 2006). Visual field-specific power increases at ipsilateralelectrodes were most pronounced in a frequency band between 11.5–14Hz for LVF trials and between 17 and 20 Hz for RVF trials (data notshown). To directly compare between visual fields, these slightly differentfrequency bands were summarized for the analyses reported below byinvestigating changes in the higher alpha–lower beta (11.5–20 Hz) fre-quency band for both LVF and RVF trials. To make sure that the observedactivity is indeed lateralized between the hemispheres and did not resultfrom a general pattern across hemispheres, significant power differencesin the alpha/beta frequency band between left (P3/PO3) and right (P4/PO4) electrode pools were identified by means of continuous Wilcoxonsigned-rank tests (10 ms time windows) for both LVF and RVF trials. Totest whether the obtained laterality differences were driven by powerdifferences between non-interference and interference conditions atelectrode sites over the ipsilateral competitor-related or the contralateral

Waldhauser et al. • Alpha/Beta Oscillations Indicate Memory Inhibition J. Neurosci., February 8, 2012 • 32(6):1953–1961 • 1955

target-related hemisphere, we additionally calculated Wilcoxon signed-rank tests for 10 ms time windows separately for the left and right elec-trode pools.

Competitor-related and target-related lateralization analysis. Based onthe visual field-specific analysis, we isolated competitor-related andtarget-related effects on alpha/beta power (11.5–20 Hz) independent ofthe specific visual field in which the respective memory items were pre-sented. Due to the similarity in lateralization and only slight differencesin timing, the visual field-specific time windows were integrated intothree time windows that were used for further analyses, capturing effectsin both LVF and RVF trials starting from the earliest onset of lateralityeffects to the latest offset in either LVF or RVF trials. To increase sensi-tivity, we summarized EEG power within the competitor and targethemispheres for the LVF and RVF trials (by flipping the topography forthe RVF condition; cf. Spitzer and Blankenburg, 2011). As a result, forboth LVF and RVF trials the activity in the competitor hemisphere wasartificially represented at the left electrodes, while activity in the targethemisphere was now reflected at right electrodes. We compared EEGpower between the interference and the non-interference conditions atright (P4/PO4) and left posterior electrode pools (P3/PO3). We alsoinvestigated the lateralization of these effects by comparing the differ-ences between the interference and the non-interference condition be-tween the two posterior electrode pools.

Lateralization effects modulated by memory status of competitor andtarget. Central to the idea of memory inhibition is the assumption thatcompetitors should only be inhibited when they are strong enough tointerfere during memory retrieval (Anderson et al., 1994; Bauml, 1998;Norman et al., 2007). That is, memory inhibition engaged to handleretrieval competition is driven by the strength of the competitor memoryrather than by the strength of the target memory (Anderson, 2003; Stormet al., 2006). EEG effects related to interference and memory inhibitionshould therefore be observable for trials where competitors have a highpotential to interfere, regardless of the strength of the target memory. Totest for this assumption, we contrasted EEG power for trials within theinterference condition for which encoding had been successful for boththe target and the competitor (fully encoded), only the competitor (com-petitor encoded), or only the target (target encoded). We expected asimilar increase in early alpha/beta power for the fully-encoded andcompetitor-encoded trials when compared to target-encoded trials. Thecomparison only included participants (n � 10) that had a reasonablehigh number of trials for which competitors or targets were not learned(�10, M � 20). The analysis focused on alpha/beta power (11.5–20 Hz)over target-related right and competitor-related left posterior electrodepools in the first time window obtained in the preceding analysis (90 –430 ms).

Effects nonspecific to hemisphere. To investigate the effects of selectivememory retrieval on neural oscillations that occurred independently ofhemisphere, we compared EEG power between the interference and thenon-interference condition, collapsing LVF and RVF trials without flip-ping the EEG channels along the midline. This was done to replicateprevious findings showing that retrieval in an interference conditionleads to higher theta power (5– 8 Hz) than retrieval in a non-interferencecondition (Hanslmayr et al., 2010). As in the comparison for the alpha/beta oscillations, we only included trials for which the cue– competitorcombinations were successfully retrieved during the study phase. Timewindows and electrodes were selected based on visual inspection of thedataset (see Fig. 5).

In a second step, analogous to the alpha/beta effects, we expected anincrease of theta power in the early (180 –500 ms) time window withhigher interference in trials where the competitor had been successfullyencoded, regardless of encoding status of the target. Based on inspectionof the data set, we compared EEG power in a narrow frequency bandaround 6.5 (� 0.2) Hz at a right frontocentral electrode pool (FC6, C4).

Brain behavior and cross-frequency correlations. To investigate thefunctional significance of the early alpha/beta increase, we tested whetherEEG power in the 90 – 430 ms time window correlated with the forgettingof competitors and facilitation of the targets. With respect to previousfindings (Hanslmayr et al., 2010), we also correlated early (180 –500 ms)and late (1200 –1800 ms) theta power with forgetting and facilitation. A

forgetting index was obtained by subtracting recall scores for competitoritems from baseline recall rates and dividing the result by baseline recallfor each subject (Kuhl et al., 2007). Positive scores indicate retrieval-induced forgetting corrected for baseline memory performance. Simi-larly, a facilitation index was obtained by subtracting baseline recall ratesfrom competitor recall rates and dividing the result by baseline recallrates. Given the suggested link between inhibition and interference de-tection in RIF (Anderson, 2003), we correlated early alpha/beta powerwith theta power in the interference condition. Correlation analyses wereconducted by calculating nonparametric Spearman’s rank correlationcoefficients (rs).

ResultsBehavioral resultsThe behavioral data from the final recall test are reported in Table1. The ANOVA revealed a significant CONDITION (interferencevs non-interference) by ITEM (competitor vs baseline) interac-tion (F(1,17) � 4.57, p � 0.05). This interaction was due to lowerrecall performance for competitor compared to baseline items inthe interference condition (t(17) � 2.157, p � 0.05). No such effectwas evident in the non-interference condition (t(17) � 1.657, p �0.10). This pattern demonstrates that our manipulation was ef-fective in causing RIF and selectively triggered inhibitorymechanisms in the interference condition. A main effect ofCONDITION emerged (F(1,17) � 32.947, p � 0.001) due to betterrecall in the non-interference condition. Notably, there was noCONDITION by ITEM by HEMISPHERE (left vs right) interac-tion (F(1,17) � 0.268, p � 0.6), indicating that RIF did not differbetween the left and right hemisphere.

To investigate whether retrieval practice also had facilitatory ef-fects on memory performance, a similar three-way ANOVA wasconducted with the factors CONDITION (interference vs non-interference), ITEM (target vs baseline), and HEMISPHERE. How-ever, no significant main effect for ITEM or any interaction withthis factor emerged (F(1,17) � 1.769, p � 0.2), showing that re-trieval practice did not lead to higher recall rates of practicedwhen compared to unpracticed items, neither within the inter-ference condition (M � 0.91, SEM � 0.02 vs M � 0.93, SEM �0.01; t(17) � 1.190, p � 0.25) nor within the non-interferencecondition (M � 0.98, SEM � 0.004 vs M � 0.97, SEM � 0.01;t(17) � 1.277, p � 0.2). These null results are most likely due toceiling effects given the high overall recall rates and will thereforenot be discussed further.

EEG resultsVisual field-specific effectsResults of the visual field-specific analysis are shown in Figure 3(gray-shaded areas), depicting stretches of significant (p � 0.05)differences between electrode pools. Magnitude of the differencesis visualized as lateralization indices obtained by subtracting al-pha/beta power at the right electrode pool from the one at the leftelectrode pool (see Fig. 3, middle row). For trials in which theretrieval cue was presented in the LVF, positive values on thislateralization index indicate higher power over the ipsilateral left

Table 1. Recall rates for baseline and competitor items by hemisphere

Hemisphere

Interference condition Non-interference condition

Baseline Competitor Baseline Competitor

Right 0.93 � 0.01 0.91 � 0.02 0.97 � 0.01 0.99 � 0.00Left 0.96 � 0.01 0.92 � 0.02 0.98 � 0.01 0.99 � 0.00Both 0.94 � 0.01 0.91 � 0.01 0.97 � 0.01 0.99 � 0.00

Note. Recall scores (M � SEM) in the final test phase for learned baseline and competitor items by condition andhemisphere.


hemisphere housing the representation of the competitor item.For RVF cues, higher power over the ipsilateral competitor-related hemisphere is indicated by negative values. The statisticalanalysis revealed three time windows in which lateralization ef-fects were significant for each LVF and RVF trial. The three timewindows showing significant laterality effects differed slightly be-tween LVF and RVF trials (LVF: 90 –330 ms, 560 –790 ms, and1600 –1770 ms; RVF: 360 – 430 ms, 570 – 810 ms, and 1670 –1800ms). In these three time windows, the power over target-relatedcontralateral electrode sites was decreased compared tocompetitor-related electrodes ipsilateral to the cue (Fig. 3, p val-ues �0.05). Early in the epoch (LVF: 80 –200 ms; RVF: 200 –260and 280 –300 ms) the laterality differences were due to higheralpha/beta power in the interference condition over the ipsilat-eral competitor-related hemisphere (see orange color bars in Fig.3; p values �0.05). Later in the epoch (LVF onset: 420 ms; RVFonset: 570 ms), laterality differences were due to an underlying

decrease in alpha/beta power for the inter-ference condition over the contralateraltarget-related hemisphere (color bars inmagenta, Fig. 3; p values �0.05).

Competitor-related andtarget-related effectsFlipping the EEG channels along the mid-line for the RVF condition and poolingtrials across LVF and RVF conditions al-lowed the assessment of competitor-relatedactivity at left hemispheric posterior elec-trodes and of target-related activity at righthemispheric posterior electrodes. The timecourse of the difference in alpha/beta po-wer between the interference and non-interference conditions for the competitorand target-related hemispheres is plotted inFigure 4B. The first effect between 90 and430 ms after stimulus onset was drivenby higher levels of alpha/beta power inthe interference condition than in thenon-interference condition over thecompetitor-related hemisphere (Z �2.373, p � 0.05; Fig. 4A, left). Over thetarget-related hemisphere, alpha/betapower was numerically decreased in the in-terference condition when compared to thenon-interference condition in this timewindow, but no significant difference be-tween the two conditions emerged (Z �1.285, p � 0.15). The alpha/beta power in-crease for the interference versus non-inter-ference condition was lateralized, asreflected by significantly stronger alpha/betapower over the competitor-related hemi-sphere than over the target-related hemi-sphere (Z � 2.461, p � 0.05). A secondeffect was obtained 560–810 ms after stim-ulus onset. When compared to the non-interference condition, alpha/beta power inthe interference condition was significantlydecreased over the target-related hemi-sphere (Z � 3.245, p � 0.005; Fig. 4A, mid-dle). No difference between the twoconditions was obtained over the competi-tor-related hemisphere (Z � 1.372, p �

0.15). The effect was lateralized; the difference between interferenceand non-interference conditions was significantly greater in thetarget-related hemisphere than in the competitor-relatedhemisphere (Z � 3.070, p � 0.005). A third effect emerged1600 –1800 ms after stimulus onset. Alpha/beta power was signif-icantly decreased for the interference condition when comparedto the non-interference condition over both hemispheres (Z val-ues �2.678, p values �0.01; Fig. 4A, right). However, the effectwas again lateralized, with a higher alpha/beta power decreaseover the target-related hemisphere than over the competitor-related hemisphere (Z � 2.286, p � 0.05).

Lateralization effects modulated by interference of the competitorBoth visual field-specific analyses and summarized competitor-related and target-related analyses suggest that competitive re-trieval in the interference condition is characterized by an earlyincrease in alpha/beta power over the competitor-related brain

Figure 3. Visual field-specific lateralization effects. Middle row, Laterality index for LVF and RVF trials. Gray-shaded areasindicate significant laterality differences in alpha/beta power between left and right electrode pools specific for each LVF and RVFtrials. Dotted boxes show the time windows selected for further statistical analyses to capture effects in both LVF and RVF trials.Colored bars at the top and bottom of the plot indicate significant differences between interference and non-interference condi-tions at the electrode pool ipsilateral (orange) and contralateral (magenta) to the retrieval cue. Top and bottom rows, Topographicmaps showing the alpha/beta power distribution over the scalp in the time windows indicated by the gray-shaded areas in themiddle row for LVF (top row) and RVF (bottom row) trials. Left and right electrode pools are depicted in white.


hemisphere. We further investigated this assumption by takinginto account that inhibition associated with RIF is dependent onthe degree of interference arising from the competitor rather thanon the strength of the target item (Anderson et al., 1994; Bauml,1998; Anderson, 2003; Norman et al., 2007). Thus, we testedwhether early alpha/beta power (90 – 430 ms) is modulated by thememory status of the competing memory trace. Only competitoritems that are encoded in memory should interfere with the re-trieval of the target. The results of this analysis revealed higheralpha/beta power for fully encoded trials over the competitor-related hemisphere when compared to target-encoded trials (Z �2.701, p � 0.01; see Fig. 4C); no significant difference between thetwo conditions was obtained over the target-related hemisphere(Z � 0.459, p � 0.6; see Fig. 4C). We also obtained a significantincrease of alpha/beta power between 90 and 430 ms over thecompetitor hemisphere for competitor-encoded trials relative totarget-encoded trials (Z � 2.497, p � 0.05). No effect was ob-tained over the target hemisphere (Z � 0.255, p � 0.75). Therewas no difference between fully encoded trials and competitor-encoded trials over either the competitor hemisphere (Z � 0.968,p � 0.3) or the target hemisphere (Z � 0.663, p � 0.5). The meanincrease of alpha/beta power for fully-encoded and competitor-encoded trials when compared to target-encoded trials was sig-

nificantly greater over the competitor hemisphere than over thetarget hemisphere (Z � 2.293, p � 0.05).

Interference effects on theta power, nonspecific to hemisphereWe examined theta power differences between the interferenceand non-interference conditions, as two previous studies indi-cated that enhanced frontal theta power indexes interference inselective memory retrieval (Hanslmayr et al., 2010; Staudigl et al.,2010). The results of this analysis are plotted in Figure 5. Repli-cating the prior studies, theta power was higher in the interfer-ence than in the non-interference condition at frontal electrodesites (FZ, FP2,and F4; see Fig. 5). This difference reached signifi-cance in two time windows, 180 –500 ms (Z � 2.112, p � 0.05)and 1200 –1800 ms (Z � 1.982, p � 0.05).

We also tested whether theta power as the presumed indi-cator of interference detection was modulated by memorystrength of the competitor. When compared to target encodedtrials (MSignal change � 7.4%, SEM � 5.4), theta power was increasedfor both fully encoded (MSignal change � 19.0%, SEM � 5.1; Z �1.988, p � 0.05) and for competitor encoded trials (MSignal change �17.6%, SEM � 4.2; Z � 2.09, p � 0.05). There was no differencebetween fully-encoded and competitor-encoded trials (Z � 0.357,p � 0.7).

Figure 4. Competitor-related and target-related alpha/beta power at posterior electrodes sites and its correlation with later forgetting. A, The competitor-related power differences betweenconditions (interference � non-interference) are shown over the left hemisphere. Target-related activity is depicted over the right hemisphere. Electrodes selected for statistical analyses aredepicted in white. B, Alpha/beta power differences between conditions (interference�non-interference) at the selected electrode pools over competitor and target hemisphere. Gray-shaded areascorrespond to the time windows shown in A. C, Mean (SEM) alpha/beta power between 90 and 430 ms for trials from the interference condition depending on whether both target and competitor(“fully encoded”), only the competitor (“competitor encoded”), or only the target (“target encoded”) were successfully learned during the study phase. Asterisk (*) indicates significant comparisons( p � 0.05); n.s., not significant; n � 10. D, Correlation between forgetting [(baseline recall � competitor recall)/baseline recall)] and alpha/beta power in the interference condition over thecompetitor hemisphere (n � 18).


Correlation between alpha/beta power, forgetting, facilitation, andtheta powerWe tested whether the lateralized effects of alpha/beta power andthe frontal theta effects predicted the amount of RIF. The earlyalpha/beta power increase in the interference condition over thecompetitor-related hemisphere between 90 and 430 ms corre-lated positively with later forgetting (rs � 0.416, p � 0.05, n � 18,one-tailed; Fig. 4D). No correlation was observed for the non-interference condition over the competitor hemisphere (rs �0.143, p � 0.25, n � 18, one-tailed), and no correlation waspresent for the interference condition over the target hemisphere(rs � 0.011, p � 0.45, n � 18, one-tailed). Theta power did notcorrelate with forgetting in either the early or late time windows(rs � 0.154, p � 0.25, n � 18, one-tailed). Facilitation correlatedwith neither forgetting (rs � �0.165, p � 0.5, n � 18, one-tailed)nor EEG power (all rs � �0.305, p � 0.1, n � 18, one-tailed).Across frequencies, we observed a nonsignificant trend for a pos-itive correlation between early alpha/beta and theta power in theinterference condition (rs � 0.333, p � 0.1, n � 18, one-tailed).

DiscussionA prominent theory of forgetting states that retrieval competitionin episodic memory is resolved by inhibitory mechanisms thatweaken competing memory representations (Ciranni and Shi-mamura, 1999; Levy and Anderson, 2002; Anderson, 2003;Spitzer et al., 2009). Our study is the first to provide direct neuralevidence of inhibition directed at interfering episodic memoryrepresentations. Adapting a visual half-field approach to studythe mechanisms underlying retrieval competition, we separatedcompetitor-related activity from target-related activity during se-lective memory retrieval. As predicted, our novel paradigm in-duced reliable RIF for memories that competed with the selectiveretrieval of desired information. In parallel, the EEG data show anincrease of posterior alpha/beta power over the hemisphere thatstored the competing memories. Importantly, this oscillatorypattern was reliably associated with the degree of interferencearising from the competing memory representation (Anderson et

al., 1994; Bauml, 1998; Norman et al., 2007). The lateralized in-crease of alpha/beta power was specific to retrieval situationswhere the selection of a target had to be achieved against theinterference of a competitor. Furthermore, the effect was depen-dent on the strength of the competing representation, as definedby whether the competitor was encoded or not but independentof the strength of the target memory representation (Anderson,2003). It is widely agreed that an increase in oscillatory alpha/betapower reflects cortical inhibition (Klimesch et al., 2007; Jensenand Mazaheri, 2010; Hanslmayr et al., 2011a). Higher alpha/betaoscillatory power is assumed to result from the rhythmic activityof inhibitory interneurons (Klimesch et al., 2007; Jensen andMazaheri, 2010); it correlates negatively with the BOLD signal intask-relevant brain areas (Hanslmayr et al., 2011b; Scheeringa etal., 2011) and signifies the filtering and attenuation of informa-tion in perception, attention, and working memory (Worden etal., 2000; Kelly et al., 2006; Hanslmayr et al., 2007; Sauseng et al.,2009b; Handel et al., 2011). The present results extend this pat-tern to the episodic memory domain and provide the first directevidence that competing visual memories are indeed inhibitedduring selective memory retrieval. Further corroborating thisclaim, an increase of alpha/beta power over the competitor hemi-sphere predicted later forgetting. Interestingly, the competitor-related alpha/beta power increase emerged in an early timewindow and preceded the decrease of alpha/beta power in thetarget-related hemisphere. This time course indicates that inhi-bition is rapidly exerted to attenuate the irrelevant competingmemory to facilitate retrieval of the relevant target memory.

Two recent fMRI study applied multivoxel pattern classifieralgorithms to investigate competition and interference resolu-tion in episodic memory (Kuhl et al., 2011; Oztekin and Badre,2011). Reduced performance of the classifier in identifying targetmemory representations (Kuhl et al., 2011) and the degree ofreactivation of competing information (Oztekin and Badre,2011) indicated higher interference and predicted the degree offorgetting. However, these studies were unable to show how in-terference resolution modulates the representations of compet-ing memories. In accordance with theories on RIF (Anderson,2003), our data give direct evidence that interference resolutioncan be achieved by the fast and efficient inhibition of competingmemories.

The lateralization effects of increased alpha/beta power overcompetitor-related areas in the present study closely mirror thefindings from several prior studies in the selective visual attentiondomain (Worden et al., 2000; Kelly et al., 2006; Thut et al., 2006;Rihs et al., 2007; Sauseng et al., 2009b; Handel et al., 2011). Asforeshadowed by theories on memory inhibition and RIF (An-derson and Spellman, 1995; Levy and Anderson, 2002; Anderson,2003), our results suggest that the neural mechanisms regulatinginformation selection in visual attention may generalize to theepisodic memory domain. Alpha/beta oscillations are generatedby frontoparietal executive control networks (Capotosto et al.,2009; Sauseng et al., 2009b; Thut et al., 2011). In line with previ-ous neuroimaging studies on RIF (Johansson et al., 2007; Kuhl etal., 2007; Wimber et al., 2009), our findings indicate that memoryinhibition is a top-down controlled process (Anderson and Spell-man, 1995).

The pattern of decreased alpha/beta power over target-relatedareas gives further evidence for cross-domain mechanisms of in-terference resolution. Decreased alpha power is considered toreflect the higher excitability of cell assemblies in sensory areas,allowing facilitated processing of target information in selectiveattention and working memory (e.g., Sauseng et al., 2005; Thut et

Figure 5. Difference in theta power between interference and non-interference conditions.Top row, Topographical maps for the time windows selected for statistical analysis. White circlesindicate electrodes selected for analysis (FZ, FP2, F4). Bottom row, Theta power for interferenceand non-interference conditions at the electrode pool indicated in the upper row. Gray shadedareas correspond to the time windows depicted in the upper row.


al., 2006; Lundqvist et al., 2011) and indicating material-specificsensory reactivation of long-term memories (Khader and Rosler,2011). A larger decrease in this frequency range has been ob-served in high load and interference conditions (Khader andRosler, 2011; Lundqvist et al., 2011). Parallel effects are evident inour data and presumably indicate the effortful reactivation ofneural populations housing sensory features of episodicmemories.

The similarity between the current alpha/beta effect and thoseobserved in studies of selective attention raises the question ofwhether the effect in the present study likewise reflects an atten-tional mechanism. Several facets of our data speak against such aninterpretation. The lateralization effects between the interferenceand non-interference conditions were obtained despite identicalsensory stimulation. The effects thus varied purely as a functionof the location where the interfering competitor memory hadbeen encoded. This makes it difficult to explain a lateralized in-crease of alpha/beta power with selective attention mechanismswithout recurring to the memory domain. It could be argued thatdifferences in visual stimulation and perceptual processing be-tween interference and non-interference conditions during en-coding could have influenced selective attention during theselective retrieval phase. However, the stability of the alpha/betaincrease across several comparisons, showing that alpha/betapower only covaried with memory strength of the competitorunder identical encoding situations, rules out such an explana-tion. Alternatively, the early alpha/beta increase signifies the en-gagement of selective attention processes that act to facilitatetarget retrieval, for example, by attenuating irrelevant visual in-put. However, such an account would predict that the EEG effectis modulated by the encoding status of the target and that itcorrelates with facilitated target recall. Neither was the case. Im-portantly, the early alpha/beta power increase correlated exclu-sively with later forgetting (not facilitation). Together, thesefindings are difficult to explain by spatial attention accounts.

The frequency range capturing the lateralization effects in thepresent study (11.5–20 Hz) overlapped with what has tradition-ally been regarded as the beta frequency band (Engel and Fries,2010). It has been suggested that beta frequencies are specificallyengaged in sensorimotor processing (Pfurtscheller et al., 1996) orin signaling the current cognitive state (Engel and Fries, 2010).Oscillations in a similar frequency range as that in our study havebeen labeled and interpreted as upper alpha activity in studies onselective attention and motivated forgetting (e.g., Worden et al.,2000; Rihs et al., 2007; Bauml et al., 2008). Alpha and beta activityco-occur during episodic memory retrieval and selective atten-tion (Zanto and Gazzaley, 2009) and emerge simultaneously inrealistic neural network models (Lundqvist et al., 2011). Also,beta oscillatory activity correlates with higher working memoryfunctioning due to more effective inhibition of distracting infor-mation (Zanto and Gazzaley, 2009) and predicts a decrease inneural activity in task-active brain regions (Hanslmayr et al.,2011b; Scheeringa et al., 2011). These findings show that the al-pha and beta frequency bands are highly related and might servesimilar neural and cognitive functions.

A number of previous studies have shown that frontal thetapower varies as a function of interference in response conflicttasks (Hanslmayr et al., 2008; Cavanagh et al., 2009) and morespecifically during episodic memory retrieval (Khader andRosler, 2011). For instance, higher levels of theta power are ob-tained during competitive retrieval when compared to noncom-petitive memory retrieval (Hanslmayr et al., 2010). Ourobservation of increased oscillatory theta power over frontal elec-

trode sites closely replicates prior findings (Hanslmayr et al.,2010; Staudigl et al., 2010). Timing differences between the pres-ent and prior studies could be due to differences in stimulusmaterial (verbal vs pictorial), experimental design (blocked vsalternating trial succession), and EEG baseline conditions. Theselective increase in theta power indicates that the interferencecondition indeed triggered higher levels of competition than thenon-interference condition. Similarly, the behavioral results in-dicate that competition was specifically elicited in the interfer-ence condition as reflected in the lower recall rates of thecompetitor compared with baseline items. Finally, our data showthat early theta and alpha/beta power are similarly triggered bythe interference potential of the competitor, and a trend for across-frequency correlation suggests that effects in the two fre-quency bands may be interrelated.

ConclusionsThe present experiment sheds new light onto how retrieval compe-tition is resolved in episodic memory. Using lateralized presentationof visual memories, we were able to disentangle competitor-relatedfrom target-related neural activity. Our results provide direct neuralevidence that retrieval competition is resolved by inhibition of sen-sory brain areas processing the interfering memory representations.Moreover, our results demonstrate that inhibition occurs quickly(starting within 100 ms after cue presentation), suggesting that inhi-bition in the episodic memory system is highly efficient. These find-ings closely mirror results from studies in the selective visualattention domain and imply that similar processes regulate retrievalcompetition in episodic memory.

ReferencesAnderson MC (2003) Rethinking interference theory: executive control and

the mechanisms of forgetting. J Mem Lang 49:415– 445.Anderson MC, Spellman BA (1995) On the status of inhibitory mechanisms

in cognition: memory retrieval as a model case. Psychol Rev 102:68 –100.Anderson MC, Bjork RA, Bjork EL (1994) Remembering can cause forget-

ting: retrieval dynamics in long-term memory. J Exp Psychol Learn MemCogn 20:1063–1087.

Bauml KH (1998) Strong items get suppressed, weak items do not: the roleof item strength in output interference. Psychon Bull Rev 5:459 – 463.

Bauml KH, Hanslmayr S, Pastotter B, Klimesch W (2008) Oscillatory cor-relates of intentional updating in episodic memory. Neuroimage41:596 – 604.

Capotosto P, Babiloni C, Romani GL, Corbetta M (2009) Frontoparietalcortex controls spatial attention through modulation of anticipatory al-pha rhythms. J Neurosci 29:5863–5872.

Cavanagh JF, Cohen MX, Allen JJ (2009) Prelude to and resolution of anerror: EEG phase synchrony reveals cognitive control dynamics duringaction monitoring. J Neurosci 29:98 –105.

Ciranni MA, Shimamura AP (1999) Retrieval-induced forgetting in epi-sodic memory. J Exp Psychol Learn Mem Cogn 25:1403–1414.

Corballis PM, Gratton G (2003) Independent control of processing strate-gies for different locations in the visual field. Biol Psychol 64:191–209.

Engel AK, Fries P (2010) Beta-band oscillations - signalling the status quo?Curr Opin Neurobiol 20:156 –165.

Eriksen BA, Eriksen CW (1974) Effects of noise letters upon identificationof a target letter in a nonsearch task. Percept Psychophys 16:143–149.

Gratton G, Corballis PM, Jain S (1997) Hemispheric organization of visualmemories. J Cogn Neurosci 9:92–104.

Groh-Bordin C, Busch NA, Herrmann CS, Zimmer HD (2007) Event-related potential repetition effects at encoding predict memory perfor-mance at test. Neuroreport 18:1905–1909.

Handel BF, Haarmeier T, Jensen O (2011) Alpha oscillations correlatewith the successful inhibition of unattended stimuli. J Cogn Neurosci23:2494 –2502.

Hanslmayr S, Aslan A, Staudigl T, Klimesch W, Herrmann CS, Bauml KH(2007) Prestimulus oscillations predict visual perception performancebetween and within subjects. Neuroimage 37:1465–1473.


Hanslmayr S, Pastotter B, Bauml KH, Gruber S, Wimber M, Klimesch W(2008) The electrophysiological dynamics of interference during thestroop task. J Cogn Neurosci 20:215–225.

Hanslmayr S, Staudigl T, Aslan A, Bauml KH (2010) Theta oscillations pre-dict the detrimental effects of memory retrieval. Cogn Affect Behav Neu-rosci 10:329 –338.

Hanslmayr S, Gross J, Klimesch W, Shapiro KL (2011a) The role of alphaoscillations in temporal attention. Brain Res Brain Res Rev 67:331–343.

Hanslmayr S, Volberg G, Wimber M, Raabe M, Greenlee MW, Bauml KH(2011b) The relationship between brain oscillations and BOLD signalduring memory formation: A combined EEG-fMRI study. J Neurosci31:15674 –15680.

Jensen O, Mazaheri A (2010) Shaping functional architecture by oscillatoryalpha activity: gating by inhibition. Front Hum Neurosci 4:186.

Johansson M, Aslan A, Bauml KH, Gabel A, Mecklinger A (2007) Whenremembering causes forgetting: Electrophysiological correlates ofretrieval-induced forgetting. Cereb Cortex 17:1335–1341.

Kelly SP, Lalor EC, Reilly RB, Foxe JJ (2006) Increases in alpha oscillatorypower reflect an active retinotopic mechanism for distracter suppressionduring sustained visuospatial attention. J Neurophysiol 95:3844 –3851.

Khader PH, Rosler F (2011) EEG power changes reflect distinct mechanismsduring long-term memory retrieval. Psychophysiology 48:362–369.

Klimesch W, Sauseng P, Hanslmayr S (2007) EEG alpha oscillations: theinhibition-timing hypothesis. Brain Res Brain Res Rev 53:63– 88.

Kuhl BA, Dudukovic NM, Kahn I, Wagner AD (2007) Decreased demandson cognitive control reveal the neural processing benefits of forgetting.Nat Neurosci 10:908 –914.

Kuhl BA, Rissman J, Chun MM, Wagner AD (2011) Fidelity of neural reac-tivation reveals competition between memories. Proc Natl Acad Sci U S A108:5903–5908.

Levy BJ, Anderson MC (2002) Inhibitory processes and the control of mem-ory retrieval. Trends Cogn Sci 6:299 –305.

Lundqvist M, Herman P, Lansner A (2011) Theta and gamma power in-creases and alpha/beta power decreases with memory load in an attractornetwork model. J Cogn Neurosci 23:3008 –3020.

Norman KA, Newman EL, Detre G (2007) A neural network model ofretrieval-induced forgetting. Psychol Rev 114:887–953.

Oztekin I, Badre D (2011) Distributed patterns of brain activity that lead toforgetting. Front Hum Neurosci 5:86.

Pfurtscheller G, Aranibar A (1977) Event-related cortical desynchroniza-tion detected by power measurements of scalp EEG. ElectroencephalogrClin Neurophysiol 42:817– 826.

Pfurtscheller G, Stancak A Jr, Neuper C (1996) Post-movement beta syn-chronization. A correlate of an idling motor area? ElectroencephalogrClin Neurophysiol 98:281–293.

Rihs TA, Michel CM, Thut G (2007) Mechanisms of selective inhibition invisual spatial attention are indexed by alpha-band EEG synchronization.Eur J Neurosci 25:603– 610.

Romei V, Brodbeck V, Michel C, Amedi A, Pascual-Leone A, Thut G(2008) Spontaneous fluctuations in posterior alpha-band EEG activ-ity reflect variability in excitability of human visual areas. Cereb Cor-tex 18:2010 –2018.

Sauseng P, Klimesch W, Doppelmayr M, Pecherstorfer T, Freunberger R,Hanslmayr S (2005) EEG alpha synchronization and functional cou-pling during top-down processing in a working memory task. Hum BrainMapp 26:148 –155.

Sauseng P, Klimesch W, Gerloff C, Hummel FC (2009a) Spontaneous lo-cally restricted EEG alpha activity determines cortical excitability in themotor cortex. Neuropsychologia 47:284 –288.

Sauseng P, Klimesch W, Heise KF, Gruber WR, Holz E, Karim AA, GlennonM, Gerloff C, Birbaumer N, Hummel FC (2009b) Brain oscillatory sub-strates of visual short-term memory capacity. Curr Biol 19:1846 –1852.

Scheeringa R, Fries P, Petersson KM, Oostenveld R, Grothe I, Norris DG,Hagoort P, Bastiaansen MC (2011) Neuronal dynamics underlyinghigh- and low-frequency EEG oscillations contribute independently tothe human BOLD signal. Neuron 69:572–583.

Spitzer B, Blankenburg F (2011) Stimulus-dependent EEG activity reflectsinternal updating of tactile working memory in humans. Proc Natl AcadSci U S A 108:8444 – 8449.

Spitzer B, Hanslmayr S, Opitz B, Mecklinger A, Bauml KH (2009) Oscilla-tory correlates of retrieval-induced forgetting in recognition memory.J Cogn Neurosci 21:976 –990.

Staudigl T, Hanslmayr S, Bauml KH (2010) Theta oscillations reflect thedynamics of interference in episodic memory retrieval. J Neurosci30:11356 –11362.

Storm BC, Bjork EL, Bjork RA, Nestojko JF (2006) Is retrieval success anecessary condition for retrieval-induced forgetting? Psychon Bull Rev13:1023–1027.

Taft C, Sivik L (1997) Salient color terms in four languages. Scand J Psychol38:29 –34.

Thut G, Nietzel A, Brandt SA, Pascual-Leone A (2006) Alpha-band electro-encephalographic activity over occipital cortex indexes visuospatial atten-tion bias and predicts visual target detection. J Neurosci 26:9494 –9502.

Thut G, Veniero D, Romei V, Miniussi C, Schyns P, Gross J (2011) Rhyth-mic TMS causes local entrainment of natural oscillatory signatures. CurrBiol 21:1176 –1185.

Wimber M, Rutschmann RM, Greenlee MW, Bauml KH (2009) Retrievalfrom episodic memory: Neural mechanisms of interference resolution.J Cogn Neurosci 21:538 –549.

Worden MS, Foxe JJ, Wang N, Simpson GV (2000) Anticipatory biasing ofvisuospatial attention indexed by retinotopically specific alpha-band elec-troencephalography increases over occipital cortex. J Neurosci 20:RC63.

Zanto TP, Gazzaley A (2009) Neural suppression of irrelevant informationunderlies optimal working memory performance. J Neurosci 29:3059 –3066.


behavioral/systems/cognitive alpha

Documents